基于FME生产典型地形要素解译样本数据  

Production of Typical Terrain Element Interpretation Sample Data Based on FME

作　　者：刘冬枝 郝利娟 LIU Dongzhi;HAO Lijuan(Chongqing Institute of Surveying and Mapping,MNR,Chongqing 401120,China;Inner Mongolia Autonomous Region Surveying and Mapping Geographic Information Center,Hohhot 010020,China)

机构地区：[1]自然资源部重庆测绘院,重庆401120 [2]内蒙古自治区测绘地理信息中心,内蒙古呼和浩特010020

出　　处：《测绘与空间地理信息》2024年第9期127-130,共4页Geomatics & Spatial Information Technology

摘　　要：遥感影像智能化自动处理技术发展相对滞后,深度学习技术显著提高了影像特征提取成效,但深度学习样本数量和类型有限。本文以地理国情普查与监测项目遥感影像解译样本数据、地表覆盖和地理国情要素成果数据为基础,基于FME软件函数库中丰富的转换器,编写工作流生产建筑物、水域、道路、植被等典型地形要素解译样本数据,用于扩充高质量典型地形要素遥感解译样本数据库,支撑多传感器、多时相、多区域的深度学习训练,提升典型地形要素自动识别与提取准确率,为解决国产卫星影像处理和典型地形要素信息提取方面的关键技术问题提供样本数据支撑。The development of intelligent automatic remote sensing image processing technology is relatively lagging behind.Deep learning technology has significantly improved the effect of image feature extraction,but the number and type of deep learning samples are limited.Based on the remote sensing image interpretation sample data,land cover,and national geographical element result data of the national geographical census and monitoring project,and based on the abundant converters in the FME software function library,the work flow is made to produce typical terrain element interpretation sample data such as buildings,water,roads,vegetation,etc.Typical terrain element interpretation sample data is used to expand the high-quality typical terrain element remote sensing interpretation sample database,support multi-sensor,multi-temporal,multi-regional deep learning training,improve the automatic recognition and extraction accuracy of typical terrain elements,and provide sample data support for solving key technical problems in domestic satellite image processing and typical terrain element information extraction.

关 键 词：典型地形要素解译样本数据 标注数据 深度学习 FME 

分 类 号：P237[天文地球—摄影测量与遥感]